Chapter IV: Changes in organization of conceptual units

Chapter overview

In this chapter, I focus on the second of my three key questions about the development of representations of mental life: How are the conceptual units that anchor representations of mental life organized in relation to each other, and how does this organization change over development? As in Chapter III, to address this question I draw on data from all of the current studies (Studies 1-4); for details about the methods of these studies, see Chapter II. The goal of this chapter is to provide “snapshots” of the organization of conceptual units in early childhood, middle childhood, and adulthood.

General analysis plan

High-level overview

My goal in this chapter is to examine the relationships among the “conceptual units” identified in Chapter III. How does a participant’s assessment of one conceptual unit for a particular target character (e.g., the degree to which he or she indicates that a beetle is capable of the physiological sensations of the BODY) affect that participant’s assessments of other conceptual units for that target character (e.g., his or her assessment of the beetle’s capaciteis in the domains of HEART or MIND)?

I focus in particular on the possibility that the mental capacity attributions documented by the studies included in this dissertation—re-analyzed as indicators of the broader “conceptual units” identified in Chapter III—might shed light on the hierarchical organization of these conceptual units, i.e., which conceptual units might be more basic or fundamental vs. more complex, and whether any of these conceptual units might or might not be considered to depend on the presence of others. In Chapter II, I illustrated this with the following example: If many participants endorse capacities associated with Conceptual Unit A without endorsing capacities associated with Conceptual Unit B, but very few participants do the reverse (endorsing capacities associated with Conceptual Unit B but not Conceptual Unit A), this provides some evidence that Conceptual Unit A is more basic or fundamental than Conceptual Unit B, or that Conceptual Unit B somehow depends on (perhaps requires) Conceptual Unit A.

Here I will translate this general interest in the relationships among conceptual units, as well as the specific intuition about how to detect the kinds of asymmetries that would be the signature of hierarchical relationships, into a specific analysis plan to be applied to each of these datasets in turn.

Details of analyses

Unlike the previous chapter, in which I employed a canonical approach to identifying latent constructs through analyses of correlation structures—exploratory factor analysis (EFA)—in this chapter there is no tried-and-true method for meeting my analysis goals. Instead, I chart my own course through these datasets, using the EFA solutions reported in Chapter II to score participants’ endorsements of each conceptual unit for the particular target character(s) that they assessed, examining visualizations of the complicated relationships among these endorsements, and then conducting more targeted statistical analyses of one particular aspect of these relationships: the possibility of asymmetries in the endorsements of these conceptual units.

Scoring endorsements of conceptual units

The first step in these analyses is to transform participants’ ratings of individual mental capacities into “scores” that indicate the extent to which they endorsed a particular conceptual unit for the target character(s) that they were assigned to assess. To do this, I make use of the EFAs presented in Chapter II—which originally served to identify a set of conceptual units in a particular sample—to a new end: the construction of “scales” for each of these conceptual units. Scale construction is a common use of EFA and similar dimensionality reduction analyses (if anything, more common than using EFA to make the kinds of theoretical arguments featured in Chapter II).

For each EFA solution, I construct a scale for each of the factors (conceptual units) identified by that solution. First, I sort each of the mental capacities included in that study into categories based on their loadings on each of the factors in that solution. For each mental capacity, I identify the “dominant” factor as the factor with the largest positive factor loading. For example, if the mental capacity feel happy had loadings of 0.60 on the BODY factor, 0.70 on the HEART factor, and 0.30 on the MIND factor, I would sort it into the HEART category. For each factor, I take the six highest-loading items as a candidate scale, then “drop” the capacities with the smallest factor loadings on their respective dominant factors until I have the same number of mental capacities in each category. For example, if the BODY factor were the dominant factor for nine mental capacities, the HEART factor for six mental capacities, and the MIND factor for five capacities, for each factor I would keep only the capacities with the five highest positive loadings on that factor, in order to construct three scales of equal length (and a maximum length of six items).

To calculate scores on these scales, I take the average of all of mental capacities for each scale. This yields a dataset in which each participant is associated with one score for each of the conceptual units identified in the relative EFA solution, for each of the target characters that that participant assessed.

In this chapter, I apply this method to all of the three-factor solutions for adult samples as presented in Chapter II (Studies 1-4), yielding BODY, HEART, and MIND scores for each target character as assessed by each participant. (I ignore the aberrant four-factor solution for adults in Study 2 suggested by one of the three factor retention protocols considered in that chapter, since this was the only study out of the seven considered in which a four-factor solution appeared to add any value beyond the robust BODY-HEART-MIND framework common to all studies. [XX APPENDIX B?])

For child samples, I apply this method both to the solutions emerging from children’s own data and to the corresponding adult solutions for the studies in question. This allows me to explore both the relationships among children’s own conceptual units, and the relationships among a fully adult-like set of conceptual units (presumably, the conceptual units that these children will eventually come to agree on as they mature into adults).

For “older” children (7-9y of age; Studies 2 and 3), this means examining two sets of three conceptual units for each sample of children: (1) BODY, HEART, and MIND as defined by adults in that study; and (2) BODY, HEART, and MIND as defined by the children themselves. Because the EFAs for older children and adults are so similar (see Chapter II), I expect the outcomes of these two approaches to constructing BODY, HEART, and MIND scales to yield very similar results in this age group.

For “younger” children, this means examining multiple sets of conceptual units for each sample of children, including (1) BODY, HEART, and MIND as defined by adults in that study; and (2) varying sets of 2-4 conceptual units as defined by EFA of children’s own responses, using different factor retention protocols. The EFAs of younger children’s responses were less reliable, and less adult-like, than those of older children. As a consequence, I expect the outcomes of these different approaches to constructing scales for each conceptual unit to yield rather different results in this age group.

The reader should note that this is far from the only way to approach “scoring” participants on these conceptual units. In particular, I could have examined factor scores—summaries of each factor (conceptual unit) based on a participant’s responses to all mental capacities and the relationships between all mental capacities and all factors included in that EFA solution—rather than constructing “scales” in the way I have just outlined. However, much like z-scores, factor scores indicate where a participant falls in relation to other participants in the sample, and do not provide the kind of absolute score that is key to my goal in this chapter, which is to analyze relationships among factors in terms of the extent to which individual participants indicated that target characters “possessed” the conceptual units BODY, HEART, and MIND. [XX APPENDIX B?]

Even within the “scale” approach described in this section, there are many parameters of this analysis that I could have set differently. For example, I could have considered absolute factor loadings rather than raw factor loadings, which would allow for mental capacities that loaded especially strongly negatively on a particular factor to contribute (negatively) to scores on that conceptual unit; I could have omitted the step of making the scales for all factors within a single EFA sotluion equal length; I could have chosen to use only the top 4 or 5 mental capacities across all EFA solutions, or to set no limit on the number of items in a scale; or I could have implemented absolute thresholds for how strongly a mental capacity must load on a factor in order to count toward the score for that conceptual unit, or absolute limits on the degree to which a mental capacity can “cross-load” on non-dominant factors and still count toward the score for any one conceptual unit. [XX APPENDIX B?] However, these kinds of details differ quite dramatically across studies and age groups. For example, in some samples there are no strong negative factor loadings, and in others there are; if I considered absolute loadings rather than raw loadings, I could end up comparing scores from a “bipolar” scale in one sample to scores from a “unipolar” scales in another sample, making the comparison more difficult to interpret. Likewise, some EFA solutions tended to feature generally weaker factor loadings than others; if I were to impose absolute thresholds for the strength of factor loadings, I could end up comparing scores from scales of wildly different lengths across samples. In my view, the analysis decisions outlined above maximize comparability across studies and age groups—the primary goal of this chapter. (Note, however, that in the analysis code for this chapter I have included easy short cuts for the interested reader to explore different options for each of these parameters.)

Visualizing relationships

XX

outline:

• for any sample, I have identified relevant conceptual units (chapter II) and scored each participant’s endorsement of each of these units for their assigned target character(s) (previous section)
• now, I want to assess the relationships among these endorsements
• but I don’t really know what to expect, and these relationships could be very complex
• so, start with holistic, exploratory visualization: scatterplots of pairs of conceptual units

Formal analyses of asymmetries

XX

outline:

• now, a more targeted approach to one particular aspect of these relationships: asymmetries
• reminder of basic intuition: in this sample, do participants tend to endorse A without B? B without A? is there an asymmetry here?
• connect to the visualizations: asymmetries of points around the y=x line of equivalence
• one way to boil this down to a number: difference scores (for each pair), comparisons to 0 via bayesian regressions
• for any pair: if 0, no evidence for hierarchy. if non-zero, some evidence.
• comparisons across pairs: further from 0, more hierarchical.
• caveat 1: these are post-hoc analyses! but still using confirmatory stats.
• caveat 2: scores of (0,0) treated same as scores of (6,6), or anywhere in between.
• maximal random effects structures (i.e., random intercepts for participants only for Studies 1c and 4)

Study 1: An adult endpoint

In the context of this dissertation, Study 1 serves the role of describing a developmental endpoint for conceptual representations of mental life. In this chapter, I focus on what these studies can reveal about the relationships among the conceptual units discussed in Chapter III. These analyses were not part of the original publication of these studies (Weisman et al., 2017).

Special notes on data processing and analysis

As noted in Chapter II, in the original analysis of these datasets responses were recoded to run from -3 to +3 before analyses (Weisman et al., 2017); in this dissertation, I maintain the 0-6 coding for comparability to Studies 2-4.

Studies 1a, 1b, and 1c shared similar methods (namely, the “edge case” approach to eliciting variability in mental capacity attributions; see Chapter II) and yielded similar results both with respect to the conceptual units revealed by EFA (Chapter III), and with respect to the organization of these conceptual units (this chapter). For this reason, I will discuss these studies as a group before moving on to Study 1d, which employed substantially diferent methods (namely, the “diverse characters” approach to eliciting variability in mental capacity attributions).

Results

Studies 1a-1c

These studies all employed the “edge case” variant of the general approach, with participants assessing the mental capacities of a beetle and/or a robot. (See Chapter II and Weisman et al., 2017, for detailed methods.)

Studies 1a and 1b employed identical methods: US adults (Study 1a: n=405; Study 1b: n=406) each assessed a single target character on 40 mental capacities. Study 1c employed very similar methods, with the exception that participants (n=200) each assessed both target characters side by side (with left-right position counterbalaned across participants).

Scale construction

For each of these three studies, following the steps described in “General analysis plan,” above, yielded BODY, HEART, and MIND scales of 6 items each, with a large degree of overlap in items across studies; see Table 4.2.

Visualization

The visualizations of relationships among scores on these BODY, HEART, and MIND scales are remarkably similar across Studies 1a-1c (see Figure 4.1, rows A-C).

BODY vs. HEART

First I will consider the relationship between BODY and HEART (Figure 4.1, leftmost column: panels A1, B1, and C1). To my eyes, the most striking features of these visualizations are that (1) there is a positive relationship between scores on the BODY and HEART scales; and (2) there are virtually no datapoints above the line of equivalence (\(y = x\), dotted diagonal line), and certainly no datapoints in upper left quadrant of these plots. Individual participants tended to endorse the mental capacity items included in the BODY scale at least as strongly, and often more strongly, than they endorsed items included in the HEART scale—in other words, that many participants attributed more BODY than HEART to the target character in question, but virtually no participants attribute more HEART than BODY. This asymmetry appears to have been driven primarily by participants’ assessments of the beetle (in red); for the robot (in blue), BODY and HEART scores may have been more similar (close to the dotted line), and were generally quite low.

BODY vs. MIND

Next I will consider the relationship between BODY and MIND (Figure 4.1, center column: panels A2, B2, and C2). Similar to the BODY vs. HEART comparison, two notable features of these visualizations are that (1) there is a positive relationship between scores on the BODY and MIND scales; and (2) there are fewer datapoints below the line of equivalence (\(y = x\), dotted diagonal line) than above it, and no datapoints in lower right quadrant of these plots. Most participants tended to endorse the mental capacity items included in the MIND scale roughly as strongly, and sometimes more strongly, than they endorsed items included in the BODY scale, while relatively few particpiants endorsed MIND items less strongly than BODY items (though this asymmetry appears to have been less extreme than the asymmetry between BODY and HEART scores documented in the previous paragraph). In this case, the asymmetry between BODY and MIND appears to have been driven primarily by participants’ assessments of the robot (in blue); for the beetle (in red), BODY and MIND scores appear to have been more similar (close to the dotted line).

HEART vs. MIND

Finally I will consider the relationship between HEART and MIND (Figure 4.1, rightmost column: panels A3, B3, and C3). Again, two features of these visualizations are particularly striking: (1) There is a positive relationship between scores on the MIND and HEART scales; and (2) there are virtually no datapoints below the line of equivalence (\(y = x\), dotted diagonal line). The asymmetry between MIND and HEART scores appears to have been particularly extreme: Almost all participants endorsed the mental capacity items included in the MIND scale more strongly than the items included in the HEART scale. In this case, this asymmetry appears to be born out for both target characters, but perhaps more exaggerated for the beetle (in red) than the robot (in blue).

Interim discussion

Across Studies 1a-1c, all three of the relationships among the conceptual units identified in Chapter III (BODY, HEART, and MIND) appear to be characterized by two features: (1) Positive relationships, such that the more strongly a participant endorsed one conceptual unit, the more strongly they tended to endorse the others; and (2) Robust asymmetries in these positive relationships, such that participants tended to endorse MIND more strongly than BODY or HEART, and HEART more strongly than MIND. Visual inspection suggests that these asymmetries were most pronounced for comparisons involving HEART, with virtually every participant across all three of these studies endorsing both BODY and MIND more strongly than HEART for both of the “edge case” characters included in these studies (a beetle and a robot).

Analysis of asymmetries

Here I provide a formal analysis of the asymmetries revealed by the visualizations in the previous section. For each pair of conceptual units (BODY vs. HEART, BODY vs. MIND, and HEART vs. MIND), I used Bayesian regression to compare difference scores between these two conceptual units to zero, controlling for differences in assessments of the two “edge cases” that were featured as target characters in these studies (a beetle vs a robot), and including maximal random effects structures (in this case, no random effects for Studies 1a and 1b, and random intercepts for participants in Study 1c). See Figure 4.1, panels A-C for visual depictions of these difference scores.

BODY vs. HEART

These regression analyses confirmed that, across Studies 1a-1c, difference scores comparing scores on the BODY and HEART scales were substantially non-zero, in the direction of participants endorsing BODY items more strongly than HEART items (see the “Intercept” row for the “BODY-HEART” comparison in Table 4.1). As I speculated in the previous section, in all studies this difference was driven by participants’ assessments of the beetle; in the aggregate, difference scores were reduced to 0 for the robot (see the “Target character” row for the “BODY-HEART” comparison in Table 4.1).

BODY vs. MIND

These regression analyses confirmed that, across Studies 1a-1c, difference scores comparing scores on the BODY and MIND scales were substantially non-zero, in the direction of participants endorsing MIND items more strongly than BODY items (see the “Intercept” row for the “BODY-MIND” comparison in Table 4.1). Again, in all studies this difference was driven by participants’ assessments of the beetle; in the aggregate, difference scores were reduced to 0 for the robot (see the “Target character” row for the “BODY-MIND” comparison in Table 4.1).

HEART vs. MIND

These regression analyses confirmed that, across Studies 1a-1c, difference scores comparing scores on the HEART and MIND scales were substantially non-zero, in the direction of participants endorsing MIND items more strongly than HEART items (see the “Intercept” row for the “HEART-MIND” comparison in Table 4.1). Again, in all studies this difference was somewhat attenuated in assessments of the robot, relative to the beetle (but not reduced to 0; see the “Target character” row for the “HEART-MIND” comparison in Table 4.1).

Interim discussion

These formal analyses of difference scores across the BODY, HEART, and MIND scales in Studies 1a-1c confirm my information observations of asymmetries described in the previous section: Across studies, participants tended to endorse MIND more strongly than BODY or HEART, and BODY more strongly than HEART.

Table 4.1: Regression analyses of difference scores for US adults in Studies 1a-1c. The table presents results from separate Bayesian regressions of each pair of conceptual units (BODY vs. HEART, BODY vs. MIND, and HEART vs. MIND). Each regression included two fixed effect parameters: (1) the intercept, which I treat as an index of the asymmetry in attributions of the two conceptual units in question; and (2) a difference between target characters, reported here as a difference between the robot and the grand mean (GM). Intercepts are highlighted in bold, because these are the primary parameters of interest for these analyses. For each parameter, the table includes the estimate (b) and a 95% credible interval for that estimate. Asterisks indicate 95% credible intervals that do not include 0.

	Study 1a			Study 1b			Study 1c
Parameter	b	95% CI		b	95% CI		b	95% CI
BODY - HEART
Intercept	0.22	[ 0.20, 0.24]	*	0.24	[ 0.22, 0.25]	*	0.24	[ 0.22, 0.26]	*
Robot vs. GM	-0.22	[-0.24, -0.20]	*	-0.22	[-0.24, -0.21]	*	-0.24	[-0.25, -0.22]	*
BODY - MIND
Intercept	-0.28	[-0.30, -0.26]	*	-0.27	[-0.29, -0.25]	*	-0.27	[-0.29, -0.25]	*
Robot vs. GM	-0.31	[-0.33, -0.28]	*	-0.28	[-0.30, -0.25]	*	-0.32	[-0.34, -0.29]	*
HEART - MIND
Intercept	-0.50	[-0.52, -0.47]	*	-0.51	[-0.54, -0.48]	*	-0.51	[-0.54, -0.49]	*
Robot vs. GM	-0.09	[-0.11, -0.06]	*	-0.05	[-0.08, -0.03]	*	-0.08	[-0.10, -0.06]	*

Study 1d

In Study 1d, 431 US adults each assessed a single target character on 40 mental capacities. Unlike Studies 1a-1c, this study employed the “many characters” variant of the general approach, in which participants were randomly assigned to assess one of the following 21 target characters: an adult, a child, an infant, a person in a persistent vegetative state, a fetus, a chimpanzee, an elephant, a dolphin, a bear, a dog, a goat, a mouse, a frog, a blue jay, a fish, a beetle, a microbe, a robot, a computer, a car, or a stapler. (See Chapter II and Weisman et al., 2017, for detailed methods.)

Scale construction

Following the steps described in “General analysis plan,” above, yielded BODY, HEART, and MIND scales of 6 items each, with a large degree of overlap in items between these scales and the scales derived from Studies 1a-1c; see Table 4.2.

Visualization

Visualizations of relationships among scores on these BODY, HEART, and MIND scales are provided in Figure 4.1, row D.

BODY vs. HEART

First I will consider the relationship between BODY and HEART (Figure 4.1, panel D1). Much as in Studies 1a-1c (rows A-C), the most striking features of this visualization are that (1) there is a positive relationship between scores on the BODY and HEART scales; and (2) there are virtually no datapoints above the line of equivalence (\(y = x\), dotted diagonal line), and certainly no datapoints in upper left quadrant. Individual participants tended to endorse the mental capacity items included in the BODY scale at least as strongly, and often more strongly, than they endorsed items included in the HEART scale—in other words, many participants attributed more BODY than HEART to the target character in question, but virtually no participants attributed more HEART than BODY.

An analysis of mean scores by target character further reveals that, in the aggregate, characters that received relatively low BODY scores (e.g., inert objects, technologies, the fetus, the person in a persistant vegetative state, and such “lower” lifeforms as a microbe) received universally low mean HEART scores, while characters that received relatively high BODY scores (e.g., “higher” lifeforms like animals and typical humans) varied in their mean HEART scores. This raises the intriguing possibility that attributions of BODY and HEART may have been govered by some sort of “threshold” model, in which attributions of any substantial amount of HEART depend on the target character having a certain degree of BODY. (This will not be explored further in the current dissertation.)

BODY vs. MIND

Next I will consider the relationship between BODY and MIND (Figure 4.1, panel D2). As in Studies 1a-1c, two notable features of this visualization are that (1) there is a positive relationship between scores on the BODY and MIND scales; and (2) there are datapoints in the upper left but not the lower right quadrants. However, while participants who assessed certain target characters (namely, the tecnologies) tended to endorse the mental capacity items included in the MIND scale roughly as strongly, and sometimes more strongly, than they endorsed items included in the BODY scale, participants who asssessed other target characters, if anything, appear to have shown the reverse pattern, endorsing MIND items slightly less strongly than BODY items. In other words, there appears to be a less consistency in the “asymmetry” betwen BODY and MIND in Study 1d than there was in Studies 1a-1c.

HEART vs. MIND

Finally I will consider the relationship between HEART and MIND (Figure 4.1, panel D1). Much as in Studies 1a-1c (rows A-C), the most striking features of this visualization are that (1) there is a positive relationship between scores on the HEART and MIND scales; and (2) there are virtually no datapoints below the line of equivalence (\(y = x\), dotted diagonal line), and certainly no datapoints in lower right quadrant. Individual participants tended to endorse the mental capacity items included in the MIND scale at least as strongly, and often more strongly, than they endorsed items included in the HEART scale—in other words, many participants attributed more MIND than HEART to the target character in question, but virtually no participants attributed more HEART than MIND.

An analysis of mean scores by target character further reveals that, in the aggregate, characters that received relatively low MIND scores (e.g., inert objects, the fetus, and such “lower” lifeforms as a microbe) received universally low mean HEART scores, while characters that received relatively high MIND scores (e.g., more sophisticated technologies as well as “higher” lifeforms like animals and typical humans) varied in their mean HEART scores. As in the BODY vs. HEART comparison discussed earlier, this raises the intriguing possibility that attributions of HEART and MIND may have been govered by some sort of “threshold” model, in which attributions of any substantial amount of HEART depend on the target character having a certain degree of MIND. (Again, this will not be explored further in the current dissertation.)

Interim discussion

In Study 1d, many of the observations described for Studies 1a-1c were upheld. In particular, the relationships between BODY vs. HEART and between MIND vs. HEART appear to be characterized by two features: (1) Positive relationships, such that the more strongly a participant endorsed one conceptual unit, the more strongly they tended to endorse the other; and (2) Robust asymmetries in these positive relationships, such that participants tended to endorse either BODY or MIND more strongly than HEART. However, the relationship between BODY vs. MIND appears to be more variable across participants and across target characters than the generally asymmetrical relationship (with participants tending to attribute more MIND than BODY) that emerged in Studies 1a-1c.

Analysis of asymmetries

Here I provide a formal analysis of the asymmetries revealed by the visualizations in the previous section. As in Studies 1a-1c, for each pair of conceptual units (BODY vs. HEART, BODY vs. MIND, and HEART vs. MIND), I used Bayesian regression to compare difference scores between these two conceptual units to zero, controlling for differences in assessments of the twenty-one “diverse characters” that were featured as target characters in these studies. See Figure 4.1, panel D, for visual depictions of these difference scores.

BODY vs. HEART

These regression analyses confirmed that in Study 1d, as in Studies 1a-1c, difference scores comparing scores on the BODY and HEART scales were substantially non-zero, in the direction of participants endorsing BODY items more strongly than HEART items (see the “Intercept” row for the “BODY-HEART” comparison in Table 4.2).

This asymmetry was more pronounced for some characters (infant, chimpanzee, elephant, dolphin, bear, goat, mouse, frog, blue jay, fish), and less pronounced for others (adult, child, person in a persistant vegetative state (PVS), microbe, robot, computer, car; see the “Target character” row for the “BODY-HEART” comparison in Table 4.2). A full discussion of the differences between target characters is beyond the scope of this chapter.

BODY vs. MIND

These regression analyses indicated that in Study 1d, in contrast to Studies 1a-1c, difference scores comparing scores on the BODY and MIND scales were only very slightly non-zero, in the direction of participants endorsing MIND items more strongly than BODY items (see the “Intercept” row for the “BODY-MIND” comparison in Table 4.2).

Again, this asymmetry was more pronounced for some characters (child, infant, fetus, chimpanzee, bear, dog, goat), and less pronounced for others (microbe, robot, computer, car; see the “Target character” row for the “BODY-MIND” comparison in Table 4.2). A full discussion of the differences between target characters is beyond the scope of this chapter.

HEART vs. MIND

These regression analyses confirmed that in Study 1d, as in Studies 1a-1c, difference scores comparing scores on the HEART and MIND scales were substantially non-zero, in the direction of participants endorsing MIND items more strongly than HEART items (see the “Intercept” row for the “HEART-MIND” comparison in Table 4.2).

Again, this asymmetry was more pronounced for some characters (adult, child, person in a persistant vegetative state (PVS), fetus, microbe, car), and less pronounced for others (infant, dolphin, bear, goat, mouse, frog, blue jay, fish, robot; see the “Target character” row for the “HEART-MIND” comparison in Table 4.2). A full discussion of the differences between target characters is beyond the scope of this chapter.

Interim discussion

These formal analyses of difference scores across the BODY, HEART, and MIND scales in Study 1d confirm my informal observations of asymmetries described in the previous section: In this study, as in Studies 1a-1c above, participants tended to endorse MIND more strongly than BODY or HEART, and BODY more strongly than HEART.

Table 4.2: Regression analyses of difference scores for US adults in Study 1d. The table presents results from separate Bayesian regressions of each pair of conceptual units (BODY vs. HEART, BODY vs. MIND, and HEART vs. MIND). Each regression included two fixed effect parameters: (1) the intercept, which I treat as an index of the asymmetry in attributions of the two conceptual units in question; and (2) a difference between target characters, reported here as a difference between each character and the grand mean (GM). Intercepts are highlighted in bold, because these are the primary parameters of interest for these analyses. For each parameter, the table includes the estimate (b) and a 95% credible interval for that estimate. Asterisks indicate 95% credible intervals that do not include 0.

	Study 1d
Parameter	b	95% CI
BODY - HEART
Intercept	0.35	[ 0.33, 0.37]	*
Adult vs. GM	-0.33	[-0.42, -0.24]	*
Child vs. GM	-0.12	[-0.21, -0.04]	*
Infant vs. GM	0.37	[ 0.28, 0.46]	*
PVS vs. GM	-0.25	[-0.34, -0.17]	*
Fetus vs. GM	-0.04	[-0.13, 0.04]
Chimpanzee vs. GM	0.10	[ 0.02, 0.19]	*
Elephant vs. GM	0.11	[ 0.03, 0.20]	*
Dolphin vs. GM	0.14	[ 0.05, 0.22]	*
Bear vs. GM	0.22	[ 0.13, 0.31]	*
Dog vs. GM	0.07	[-0.01, 0.15]
Goat vs. GM	0.23	[ 0.15, 0.32]	*
Mouse vs. GM	0.28	[ 0.19, 0.38]	*
Frog vs. GM	0.31	[ 0.22, 0.39]	*
Blue jay vs. GM	0.30	[ 0.21, 0.39]	*
Fish vs. GM	0.20	[ 0.11, 0.29]	*
Beetle vs. GM	0.05	[-0.04, 0.14]
Microbe vs. GM	-0.21	[-0.30, -0.12]	*
Robot vs. GM	-0.39	[-0.47, -0.30]	*
Computer vs. GM	-0.36	[-0.44, -0.27]	*
Car vs. GM	-0.35	[-0.43, -0.26]	*
BODY - MIND
Intercept	-0.02	[-0.04, -0.01]	*
Adult vs. GM	0.05	[-0.02, 0.11]
Child vs. GM	0.13	[ 0.06, 0.20]	*
Infant vs. GM	0.26	[ 0.19, 0.33]	*
PVS vs. GM	0.05	[-0.02, 0.12]
Fetus vs. GM	0.11	[ 0.04, 0.18]	*
Chimpanzee vs. GM	0.11	[ 0.04, 0.18]	*
Elephant vs. GM	0.03	[-0.03, 0.10]
Dolphin vs. GM	0.03	[-0.04, 0.10]
Bear vs. GM	0.07	[ 0.00, 0.14]	*
Dog vs. GM	0.12	[ 0.06, 0.18]	*
Goat vs. GM	0.12	[ 0.05, 0.19]	*
Mouse vs. GM	0.07	[-0.01, 0.14]
Frog vs. GM	0.07	[ 0.00, 0.13]
Blue jay vs. GM	0.04	[-0.03, 0.10]
Fish vs. GM	0.03	[-0.04, 0.10]
Beetle vs. GM	0.00	[-0.07, 0.07]
Microbe vs. GM	-0.08	[-0.15, -0.01]	*
Robot vs. GM	-0.65	[-0.72, -0.58]	*
Computer vs. GM	-0.40	[-0.47, -0.34]	*
Car vs. GM	-0.18	[-0.24, -0.12]	*
HEART - MIND
Intercept	-0.38	[-0.40, -0.35]	*
Adult vs. GM	0.38	[ 0.28, 0.47]	*
Child vs. GM	0.25	[ 0.15, 0.35]	*
Infant vs. GM	-0.12	[-0.21, -0.02]	*
PVS vs. GM	0.30	[ 0.21, 0.39]	*
Fetus vs. GM	0.15	[ 0.06, 0.26]	*
Chimpanzee vs. GM	0.01	[-0.09, 0.10]
Elephant vs. GM	-0.08	[-0.17, 0.02]
Dolphin vs. GM	-0.11	[-0.20, -0.02]	*
Bear vs. GM	-0.15	[-0.24, -0.05]	*
Dog vs. GM	0.05	[-0.04, 0.13]
Goat vs. GM	-0.11	[-0.20, -0.02]	*
Mouse vs. GM	-0.21	[-0.32, -0.11]	*
Frog vs. GM	-0.24	[-0.34, -0.14]	*
Blue jay vs. GM	-0.27	[-0.36, -0.18]	*
Fish vs. GM	-0.17	[-0.27, -0.08]	*
Beetle vs. GM	-0.05	[-0.14, 0.05]
Microbe vs. GM	0.13	[ 0.03, 0.22]	*
Robot vs. GM	-0.27	[-0.36, -0.18]	*
Computer vs. GM	-0.05	[-0.14, 0.05]
Car vs. GM	0.17	[ 0.08, 0.26]	*

Discussion

XX

Table 4.3: Scales for each of the conceptual units (factors) identified by EFA for US Adults in Studies 1a-1d (see Chapter III). A checkmark indicates that a mental capacity was included in a scale for a particular study.

Capacity	Study 1a	Study 1b	Study 1c	Study 1d
BODY scale
getting hungry	✓	✓	✓	✓
experiencing pain	✓	✓	✓	✓
feeling tired	✓	✓	✓	✓
experiencing fear	✓	✓	✓	✓
experiencing pleasure	✓	✓	✓	✓
having free will	✓
being conscious		✓
having desires			✓
feeling calm				✓
HEART scale
feeling embarrassed	✓	✓	✓	✓
experiencing pride	✓	✓	✓	✓
feeling love	✓	✓	✓
experiencing guilt	✓	✓	✓	✓
holding beliefs	✓			✓
feeling disrespected	✓	✓	✓	✓
feeling depressed		✓	✓
telling right from wrong				✓
MIND scale
remembering things	✓	✓	✓	✓
recognizing someone	✓		✓
sensing temperatures	✓		✓	✓
communicating with others	✓	✓	✓	✓
seeing things	✓	✓		✓
perceiving depth	✓		✓	✓
detecting sounds		✓	✓	✓
working toward a goal		✓
making choices		✓

Study 2: Conceptual change between middle childhood (7-9y) and adulthood

In Study 2, 200 US adults and 200 US children between the ages of 7.01-9.99 years (median: 8.31y) each assessed a single target character on 40 mental capacities. To make the study appropriate for children in this age range, the wording of some the 40 mental capacities employed in Study 1 was modified to use more age-appropriate vocabulary, and participants responded on a 3-point scale (“no,” coded as 0; “kinda,” coded as 0.5, “yes,” coded as 1). This study employed the “edge case” variant of the general approach, with participants randomly assigned to assess either a beetle or a robot. (See Chapter II for detailed methods.)

Special notes on data processing and analysis

Results

Adults

Scale construction

Following the steps described in “General analysis plan,” above, yielded BODY, HEART, and MIND scales of 6 items each; see Table 4.4.

Visualization

Analysis of asymmetries

XX

Children (7-9y)

Visualization

XX

Analysis of asymmetries

XX

Developmental comparison

Children (7-9y), using children’s own scales

Scale construction

Following the steps described in “General analysis plan,” above, yielded BODY, HEART, and MIND scales of 6 items each; see Table 4.4.

Visualization

Analysis of asymmetries

XX

Discussion

XX

Study 3: Conceptual change over early and middle childhood (4-9y)

In Study 3, 116 US adults, 125 “older” children (7.08-9.98 years; median: 8.56y), and 124 “younger” children (4-6.98 years; median: 5.03y) each assessed a single target character on 20 mental capacities. To make the study appropriate for children in this age range, participants assessed a subset of the 40 mental capacities employed in Study 2, chosen to represent the three “conceptual units” revealed by Studies 1-2 (BODY, HEART, and MIND) and to cover a similar range of mental capacities as Studies 1-2. As in Study 2, participants responded on a 3-point scale (“no,” coded as 0; “kinda,” coded as 0.5, “yes,” coded as 1). This study employed the “diverse characters” variant of the general approach, with participants randomly or pseudo-randomly assigned to assess either one of the following 9 characters: an elephant, a goat, a mouse, a bird, a beetle, a teddy bear, a doll, a robot, or a computer. (See Chapter II for detailed methods.)

Special notes on data processing and analysis

Results

Adults

Scale construction

Following the steps described in “General analysis plan,” above, yielded BODY, HEART, and MIND scales of 6 items each; see Table 4.4.

Visualization

XX

Analysis of asymmetries

XX

Children (7-9y)

Visualization

XX

Analysis of asymmetries

Children (4-6y)

Visualization

Analysis of asymmetries

Developmental comparison

Discussion

XX

Study 4: A focus on early childhood (4-5y)

In Study 4, 104 US adults and 43 US children between the ages of 4.02-5.59 years (median: 4.73y) each assessed two target characters on 18 mental capacities. To make the study appropriate for children in this age range, this study employed a new set of 18 mental capacities (some but not all of which were used in Studies 1-3). In addition, participants were presented with a more child-friendly visual representation of the 3-point response scale (“no,” coded as 0; “kinda,” coded as 0.5, “yes,” coded as 1). This study employed the “edge case” variant of the general approach, with participants assessing both a beetle or a robot in sequence (with order counterbalanced across participants). (See Chapter II for detailed methods.)

Special notes on data processing and analysis

Results

Adults

Scale construction

Following the steps described in “General analysis plan,” above, yielded BODY, HEART, and MIND scales of 5 items each; see Table 4.4.

Visualization

XX

Analysis of asymmetries

XX

Children (4-5y)

Visualization

Analysis of asymmetries

Developmental comparison

Discussion

XX

Table 4.4: Scales for each of the conceptual units (factors) identified by EFA for US Adults in Studies 2-4 and for 7- to 9-year-old childrenin Studies 2 and 3. (See Appendix B for alternative scales based on younger children's EFA results, for Studies 3 and 4.) A checkmark indicates that a mental capacity was included in a scale for a particular sample.

	Study 2		Study 3		Study 4
Capacity	Adults	Children, 7-9y	Adults	Children, 7-9y	Adults
BODY scale
get/feel hungry	✓	✓	✓	✓	✓
feel pain	✓	✓	✓	✓
feel/get scared	✓	✓	✓	✓
feel tired	✓	✓	✓	✓	✓
feel safe	✓
smell things	✓	✓	✓	✓	✓
get/feel sick[...]		✓	✓		✓
get thirsty					✓
get angry				✓
HEART scale
feel proud	✓	✓	✓	✓
feel joy	✓	✓
feel/get sad	✓	✓	✓	✓	✓
feel happy	✓	✓
feel love/love someone	✓	✓	✓	✓	✓
feel guilty/sorry	✓		✓	✓	✓
get hurt feelings		✓	✓	✓
feel embarrassed			✓	✓
hate someone					✓
get lonely					✓
MIND scale
figure out how to do things/figure things out	✓	✓	✓	✓	✓
make choices	✓		✓	✓
recognize somebody else	✓
sense...far away	✓	✓	✓	✓
remember things	✓	✓	✓	✓	✓
see [things]	✓
be aware of itself		✓
be aware of things		✓	✓	✓
sense temperatures		✓	✓	✓
know stuff					✓
have thoughts/think					✓
hear [sounds]					✓

General discussion

XX

Chapter conclusion

In this chapter, I explored a second aspect of conceptual representations of mental life among US children and adults: The relationships among conceptual units. Studies 2-4 are consistent with the following theory: XX.

As in Chapter III, I urge the reader to remember that this is not the only possible interpretation of the pattern of results presented here; additional studies—in particular, studies designed to test the hypothesis that XX— could provide converging evidence or could challenge this theoretical interpretation. Instead, the primary role of the re-analysis discussed here has been to inspire the hypothesis stated in the previous paragraph and to the foundation for future tests of this hypothesis, in turn refining a general theory of this aspect of conceptual development.

In the next chapter, I apply the same exploratory spirit to the third and final aspect of conceptual representations of mental life: the application or deployment of these conceptual units in reasoning about various kinds of beings.

TO DO

• implement overall plan
• study 1: adults for each sample = 4 regs total
• study 2: adults for each sample (2); children’s for child sample (1); comparison (1) = 4 regs total
• study 3: adults for each sample (3); comparison (1) = 4 regs total
• study 4: adults for each sample (2); comparison (1) = 3 regs total
• make regression tables
• study 1 (4 regs: studies 1a-1d)
• study 2 (3 regs: adult scales for adults, children; child scales)
• study 2 (1 reg: developmental comparison)
• study 3 (3 regs: adult scales for adults, older, young children)
• study 3 (1 reg: developmental comparison)
• study 4 (2 regs: adult scles for adults, children)
• study 4 (1 reg: developmental comparison)
• move to appendix
• study 2: adults’ 4-factor solution for adults? for 7-9yo?
• study 3: 7-9yo for 7-9yo (1); 4-6yo for 4-6yo x 2 (2)
• study 4: 4-6yo for 4-6yo x 2 (2)
• reconsider adding to appendix
• study 3: 7-9yo for 4-6yo?
• all studies: comparisons across pairs of units (instead of separate regs for each unit)?